This project aims at significant improvement of specificity of dynamic MRI in detecting and diagnosing breast cancer. Although current dynamic breast MRI has high sensitivity in detecting enhancement of breast cancer, it has limited specificity in cancer diagnosis. This is partially due to the limitations of the current registration and segmentation methods used for enhancement segmentation in the dynamic breast MRI. It is worth noting that the accuracy of image registration and segmentation affects not only the segmentation of contrast enhancement, but also eventually the classification of contrast enhancement as benign or malignant, as having been proved in the literature. Better image registration and segmentation allow more accurate segmentation of foci of enhancement and in theory reduce false negatives. On the other hand, the low specificity of current dynamic breast MRI is also related to the limitation of current enhancement classification methods which use only the dynamic kinetic parameters or morphologic features, or their simple combinations for cancer diagnosis. More importantly, according to our knowledge, almost all cancer diagnosis methods were designed for classification of only mass enhancement, not the non-mass enhancement. The goal of this project is to overcome these limitations by developing two novel modules for enhancement segmentation and classification, respectively. In Aim 1, an advanced enhancement segmentation module will be developed for segmenting potentially suspicious enhancement with high sensitivity and specificity. In particular, a novel spatiotemporal registration method will be developed for consistent estimation of patient motion during the image acquisition, and also a graph-cut based segmentation technique will be developed for adaptive segmentation of enhancement regions. In Aim 2, a novel enhancement classification module will be developed to differentiate benign enhancement from malignant enhancement for both mass and non-mass enhancement cases by considering their respective enhancement morphologies. In particular, the evolving pattern of tumor architecture will be completely captured from dynamic MRI sequence and used for enhancement classification, instead of using only simple region-wise dynamic features as popularly used in the literature. The efficacy of these two modules will be evaluated by an existing database with MRI scans from over 700 patients. PUBLIC HEALTH RELEVANCE: This project aims at significant improvement of specificity of dynamic MRI in detecting and diagnosing breast cancer. In particular, an advanced enhancement segmentation module will be developed for segmenting potentially suspicious enhancement with high sensitivity and specificity. And, a novel enhancement classification module will be developed to differentiate benign from malignant enhancement for both mass and non- mass enhancement cases by considering their respective enhancement morphologies.